?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
powerpc/mm/4k: don't allocate larger pmd page table for 4k
[linux/fpc-iii.git] / tools / perf / util / intel-bts.c
blob6c2eb5da4afc0d290786881ebd9a7d9c5f83775f
1 /*
2 * intel-bts.c: Intel Processor Trace support
3 * Copyright (c) 2013-2015, Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 */
15
16 #include <endian.h>
17 #include <byteswap.h>
18 #include <linux/kernel.h>
19 #include <linux/types.h>
20 #include <linux/bitops.h>
21 #include <linux/log2.h>
22
23 #include "cpumap.h"
24 #include "color.h"
25 #include "evsel.h"
26 #include "evlist.h"
27 #include "machine.h"
28 #include "session.h"
29 #include "util.h"
30 #include "thread.h"
31 #include "thread-stack.h"
32 #include "debug.h"
33 #include "tsc.h"
34 #include "auxtrace.h"
35 #include "intel-pt-decoder/intel-pt-insn-decoder.h"
36 #include "intel-bts.h"
37
38 #define MAX_TIMESTAMP (~0ULL)
39
40 #define INTEL_BTS_ERR_NOINSN 5
41 #define INTEL_BTS_ERR_LOST 9
42
43 #if __BYTE_ORDER == __BIG_ENDIAN
44 #define le64_to_cpu bswap_64
45 #else
46 #define le64_to_cpu
47 #endif
48
49 struct intel_bts {
50 struct auxtrace auxtrace;
51 struct auxtrace_queues queues;
52 struct auxtrace_heap heap;
53 u32 auxtrace_type;
54 struct perf_session *session;
55 struct machine *machine;
56 bool sampling_mode;
57 bool snapshot_mode;
58 bool data_queued;
59 u32 pmu_type;
60 struct perf_tsc_conversion tc;
61 bool cap_user_time_zero;
62 struct itrace_synth_opts synth_opts;
63 bool sample_branches;
64 u32 branches_filter;
65 u64 branches_sample_type;
66 u64 branches_id;
67 size_t branches_event_size;
68 bool synth_needs_swap;
69 unsigned long num_events;
70 };
71
72 struct intel_bts_queue {
73 struct intel_bts *bts;
74 unsigned int queue_nr;
75 struct auxtrace_buffer *buffer;
76 bool on_heap;
77 bool done;
78 pid_t pid;
79 pid_t tid;
80 int cpu;
81 u64 time;
82 struct intel_pt_insn intel_pt_insn;
83 u32 sample_flags;
84 };
85
86 struct branch {
87 u64 from;
88 u64 to;
89 u64 misc;
90 };
91
92 static void intel_bts_dump(struct intel_bts *bts __maybe_unused,
93 unsigned char *buf, size_t len)
94 {
95 struct branch *branch;
96 size_t i, pos = 0, br_sz = sizeof(struct branch), sz;
97 const char *color = PERF_COLOR_BLUE;
98
99 color_fprintf(stdout, color,
100 ". ... Intel BTS data: size %zu bytes\n",
101 len);
102
103 while (len) {
104 if (len >= br_sz)
105 sz = br_sz;
106 else
107 sz = len;
108 printf(".");
109 color_fprintf(stdout, color, " %08x: ", pos);
110 for (i = 0; i < sz; i++)
111 color_fprintf(stdout, color, " %02x", buf[i]);
112 for (; i < br_sz; i++)
113 color_fprintf(stdout, color, " ");
114 if (len >= br_sz) {
115 branch = (struct branch *)buf;
116 color_fprintf(stdout, color, " %"PRIx64" -> %"PRIx64" %s\n",
117 le64_to_cpu(branch->from),
118 le64_to_cpu(branch->to),
119 le64_to_cpu(branch->misc) & 0x10 ?
120 "pred" : "miss");
121 } else {
122 color_fprintf(stdout, color, " Bad record!\n");
123 }
124 pos += sz;
125 buf += sz;
126 len -= sz;
127 }
128 }
129
130 static void intel_bts_dump_event(struct intel_bts *bts, unsigned char *buf,
131 size_t len)
132 {
133 printf(".\n");
134 intel_bts_dump(bts, buf, len);
135 }
136
137 static int intel_bts_lost(struct intel_bts *bts, struct perf_sample *sample)
138 {
139 union perf_event event;
140 int err;
141
142 auxtrace_synth_error(&event.auxtrace_error, PERF_AUXTRACE_ERROR_ITRACE,
143 INTEL_BTS_ERR_LOST, sample->cpu, sample->pid,
144 sample->tid, 0, "Lost trace data");
145
146 err = perf_session__deliver_synth_event(bts->session, &event, NULL);
147 if (err)
148 pr_err("Intel BTS: failed to deliver error event, error %d\n",
149 err);
150
151 return err;
152 }
153
154 static struct intel_bts_queue *intel_bts_alloc_queue(struct intel_bts *bts,
155 unsigned int queue_nr)
156 {
157 struct intel_bts_queue *btsq;
158
159 btsq = zalloc(sizeof(struct intel_bts_queue));
160 if (!btsq)
161 return NULL;
162
163 btsq->bts = bts;
164 btsq->queue_nr = queue_nr;
165 btsq->pid = -1;
166 btsq->tid = -1;
167 btsq->cpu = -1;
168
169 return btsq;
170 }
171
172 static int intel_bts_setup_queue(struct intel_bts *bts,
173 struct auxtrace_queue *queue,
174 unsigned int queue_nr)
175 {
176 struct intel_bts_queue *btsq = queue->priv;
177
178 if (list_empty(&queue->head))
179 return 0;
180
181 if (!btsq) {
182 btsq = intel_bts_alloc_queue(bts, queue_nr);
183 if (!btsq)
184 return -ENOMEM;
185 queue->priv = btsq;
186
187 if (queue->cpu != -1)
188 btsq->cpu = queue->cpu;
189 btsq->tid = queue->tid;
190 }
191
192 if (bts->sampling_mode)
193 return 0;
194
195 if (!btsq->on_heap && !btsq->buffer) {
196 int ret;
197
198 btsq->buffer = auxtrace_buffer__next(queue, NULL);
199 if (!btsq->buffer)
200 return 0;
201
202 ret = auxtrace_heap__add(&bts->heap, queue_nr,
203 btsq->buffer->reference);
204 if (ret)
205 return ret;
206 btsq->on_heap = true;
207 }
208
209 return 0;
210 }
211
212 static int intel_bts_setup_queues(struct intel_bts *bts)
213 {
214 unsigned int i;
215 int ret;
216
217 for (i = 0; i < bts->queues.nr_queues; i++) {
218 ret = intel_bts_setup_queue(bts, &bts->queues.queue_array[i],
219 i);
220 if (ret)
221 return ret;
222 }
223 return 0;
224 }
225
226 static inline int intel_bts_update_queues(struct intel_bts *bts)
227 {
228 if (bts->queues.new_data) {
229 bts->queues.new_data = false;
230 return intel_bts_setup_queues(bts);
231 }
232 return 0;
233 }
234
235 static unsigned char *intel_bts_find_overlap(unsigned char *buf_a, size_t len_a,
236 unsigned char *buf_b, size_t len_b)
237 {
238 size_t offs, len;
239
240 if (len_a > len_b)
241 offs = len_a - len_b;
242 else
243 offs = 0;
244
245 for (; offs < len_a; offs += sizeof(struct branch)) {
246 len = len_a - offs;
247 if (!memcmp(buf_a + offs, buf_b, len))
248 return buf_b + len;
249 }
250
251 return buf_b;
252 }
253
254 static int intel_bts_do_fix_overlap(struct auxtrace_queue *queue,
255 struct auxtrace_buffer *b)
256 {
257 struct auxtrace_buffer *a;
258 void *start;
259
260 if (b->list.prev == &queue->head)
261 return 0;
262 a = list_entry(b->list.prev, struct auxtrace_buffer, list);
263 start = intel_bts_find_overlap(a->data, a->size, b->data, b->size);
264 if (!start)
265 return -EINVAL;
266 b->use_size = b->data + b->size - start;
267 b->use_data = start;
268 return 0;
269 }
270
271 static int intel_bts_synth_branch_sample(struct intel_bts_queue *btsq,
272 struct branch *branch)
273 {
274 int ret;
275 struct intel_bts *bts = btsq->bts;
276 union perf_event event;
277 struct perf_sample sample = { .ip = 0, };
278
279 if (bts->synth_opts.initial_skip &&
280 bts->num_events++ <= bts->synth_opts.initial_skip)
281 return 0;
282
283 event.sample.header.type = PERF_RECORD_SAMPLE;
284 event.sample.header.misc = PERF_RECORD_MISC_USER;
285 event.sample.header.size = sizeof(struct perf_event_header);
286
287 sample.cpumode = PERF_RECORD_MISC_USER;
288 sample.ip = le64_to_cpu(branch->from);
289 sample.pid = btsq->pid;
290 sample.tid = btsq->tid;
291 sample.addr = le64_to_cpu(branch->to);
292 sample.id = btsq->bts->branches_id;
293 sample.stream_id = btsq->bts->branches_id;
294 sample.period = 1;
295 sample.cpu = btsq->cpu;
296 sample.flags = btsq->sample_flags;
297 sample.insn_len = btsq->intel_pt_insn.length;
298 memcpy(sample.insn, btsq->intel_pt_insn.buf, INTEL_PT_INSN_BUF_SZ);
299
300 if (bts->synth_opts.inject) {
301 event.sample.header.size = bts->branches_event_size;
302 ret = perf_event__synthesize_sample(&event,
303 bts->branches_sample_type,
304 0, &sample,
305 bts->synth_needs_swap);
306 if (ret)
307 return ret;
308 }
309
310 ret = perf_session__deliver_synth_event(bts->session, &event, &sample);
311 if (ret)
312 pr_err("Intel BTS: failed to deliver branch event, error %d\n",
313 ret);
314
315 return ret;
316 }
317
318 static int intel_bts_get_next_insn(struct intel_bts_queue *btsq, u64 ip)
319 {
320 struct machine *machine = btsq->bts->machine;
321 struct thread *thread;
322 struct addr_location al;
323 unsigned char buf[INTEL_PT_INSN_BUF_SZ];
324 ssize_t len;
325 int x86_64;
326 uint8_t cpumode;
327 int err = -1;
328
329 if (machine__kernel_ip(machine, ip))
330 cpumode = PERF_RECORD_MISC_KERNEL;
331 else
332 cpumode = PERF_RECORD_MISC_USER;
333
334 thread = machine__find_thread(machine, -1, btsq->tid);
335 if (!thread)
336 return -1;
337
338 thread__find_addr_map(thread, cpumode, MAP__FUNCTION, ip, &al);
339 if (!al.map || !al.map->dso)
340 goto out_put;
341
342 len = dso__data_read_addr(al.map->dso, al.map, machine, ip, buf,
343 INTEL_PT_INSN_BUF_SZ);
344 if (len <= 0)
345 goto out_put;
346
347 /* Load maps to ensure dso->is_64_bit has been updated */
348 map__load(al.map);
349
350 x86_64 = al.map->dso->is_64_bit;
351
352 if (intel_pt_get_insn(buf, len, x86_64, &btsq->intel_pt_insn))
353 goto out_put;
354
355 err = 0;
356 out_put:
357 thread__put(thread);
358 return err;
359 }
360
361 static int intel_bts_synth_error(struct intel_bts *bts, int cpu, pid_t pid,
362 pid_t tid, u64 ip)
363 {
364 union perf_event event;
365 int err;
366
367 auxtrace_synth_error(&event.auxtrace_error, PERF_AUXTRACE_ERROR_ITRACE,
368 INTEL_BTS_ERR_NOINSN, cpu, pid, tid, ip,
369 "Failed to get instruction");
370
371 err = perf_session__deliver_synth_event(bts->session, &event, NULL);
372 if (err)
373 pr_err("Intel BTS: failed to deliver error event, error %d\n",
374 err);
375
376 return err;
377 }
378
379 static int intel_bts_get_branch_type(struct intel_bts_queue *btsq,
380 struct branch *branch)
381 {
382 int err;
383
384 if (!branch->from) {
385 if (branch->to)
386 btsq->sample_flags = PERF_IP_FLAG_BRANCH |
387 PERF_IP_FLAG_TRACE_BEGIN;
388 else
389 btsq->sample_flags = 0;
390 btsq->intel_pt_insn.length = 0;
391 } else if (!branch->to) {
392 btsq->sample_flags = PERF_IP_FLAG_BRANCH |
393 PERF_IP_FLAG_TRACE_END;
394 btsq->intel_pt_insn.length = 0;
395 } else {
396 err = intel_bts_get_next_insn(btsq, branch->from);
397 if (err) {
398 btsq->sample_flags = 0;
399 btsq->intel_pt_insn.length = 0;
400 if (!btsq->bts->synth_opts.errors)
401 return 0;
402 err = intel_bts_synth_error(btsq->bts, btsq->cpu,
403 btsq->pid, btsq->tid,
404 branch->from);
405 return err;
406 }
407 btsq->sample_flags = intel_pt_insn_type(btsq->intel_pt_insn.op);
408 /* Check for an async branch into the kernel */
409 if (!machine__kernel_ip(btsq->bts->machine, branch->from) &&
410 machine__kernel_ip(btsq->bts->machine, branch->to) &&
411 btsq->sample_flags != (PERF_IP_FLAG_BRANCH |
412 PERF_IP_FLAG_CALL |
413 PERF_IP_FLAG_SYSCALLRET))
414 btsq->sample_flags = PERF_IP_FLAG_BRANCH |
415 PERF_IP_FLAG_CALL |
416 PERF_IP_FLAG_ASYNC |
417 PERF_IP_FLAG_INTERRUPT;
418 }
419
420 return 0;
421 }
422
423 static int intel_bts_process_buffer(struct intel_bts_queue *btsq,
424 struct auxtrace_buffer *buffer,
425 struct thread *thread)
426 {
427 struct branch *branch;
428 size_t sz, bsz = sizeof(struct branch);
429 u32 filter = btsq->bts->branches_filter;
430 int err = 0;
431
432 if (buffer->use_data) {
433 sz = buffer->use_size;
434 branch = buffer->use_data;
435 } else {
436 sz = buffer->size;
437 branch = buffer->data;
438 }
439
440 if (!btsq->bts->sample_branches)
441 return 0;
442
443 for (; sz > bsz; branch += 1, sz -= bsz) {
444 if (!branch->from && !branch->to)
445 continue;
446 intel_bts_get_branch_type(btsq, branch);
447 if (btsq->bts->synth_opts.thread_stack)
448 thread_stack__event(thread, btsq->sample_flags,
449 le64_to_cpu(branch->from),
450 le64_to_cpu(branch->to),
451 btsq->intel_pt_insn.length,
452 buffer->buffer_nr + 1);
453 if (filter && !(filter & btsq->sample_flags))
454 continue;
455 err = intel_bts_synth_branch_sample(btsq, branch);
456 if (err)
457 break;
458 }
459 return err;
460 }
461
462 static int intel_bts_process_queue(struct intel_bts_queue *btsq, u64 *timestamp)
463 {
464 struct auxtrace_buffer *buffer = btsq->buffer, *old_buffer = buffer;
465 struct auxtrace_queue *queue;
466 struct thread *thread;
467 int err;
468
469 if (btsq->done)
470 return 1;
471
472 if (btsq->pid == -1) {
473 thread = machine__find_thread(btsq->bts->machine, -1,
474 btsq->tid);
475 if (thread)
476 btsq->pid = thread->pid_;
477 } else {
478 thread = machine__findnew_thread(btsq->bts->machine, btsq->pid,
479 btsq->tid);
480 }
481
482 queue = &btsq->bts->queues.queue_array[btsq->queue_nr];
483
484 if (!buffer)
485 buffer = auxtrace_buffer__next(queue, NULL);
486
487 if (!buffer) {
488 if (!btsq->bts->sampling_mode)
489 btsq->done = 1;
490 err = 1;
491 goto out_put;
492 }
493
494 /* Currently there is no support for split buffers */
495 if (buffer->consecutive) {
496 err = -EINVAL;
497 goto out_put;
498 }
499
500 if (!buffer->data) {
501 int fd = perf_data_file__fd(btsq->bts->session->file);
502
503 buffer->data = auxtrace_buffer__get_data(buffer, fd);
504 if (!buffer->data) {
505 err = -ENOMEM;
506 goto out_put;
507 }
508 }
509
510 if (btsq->bts->snapshot_mode && !buffer->consecutive &&
511 intel_bts_do_fix_overlap(queue, buffer)) {
512 err = -ENOMEM;
513 goto out_put;
514 }
515
516 if (!btsq->bts->synth_opts.callchain &&
517 !btsq->bts->synth_opts.thread_stack && thread &&
518 (!old_buffer || btsq->bts->sampling_mode ||
519 (btsq->bts->snapshot_mode && !buffer->consecutive)))
520 thread_stack__set_trace_nr(thread, buffer->buffer_nr + 1);
521
522 err = intel_bts_process_buffer(btsq, buffer, thread);
523
524 auxtrace_buffer__drop_data(buffer);
525
526 btsq->buffer = auxtrace_buffer__next(queue, buffer);
527 if (btsq->buffer) {
528 if (timestamp)
529 *timestamp = btsq->buffer->reference;
530 } else {
531 if (!btsq->bts->sampling_mode)
532 btsq->done = 1;
533 }
534 out_put:
535 thread__put(thread);
536 return err;
537 }
538
539 static int intel_bts_flush_queue(struct intel_bts_queue *btsq)
540 {
541 u64 ts = 0;
542 int ret;
543
544 while (1) {
545 ret = intel_bts_process_queue(btsq, &ts);
546 if (ret < 0)
547 return ret;
548 if (ret)
549 break;
550 }
551 return 0;
552 }
553
554 static int intel_bts_process_tid_exit(struct intel_bts *bts, pid_t tid)
555 {
556 struct auxtrace_queues *queues = &bts->queues;
557 unsigned int i;
558
559 for (i = 0; i < queues->nr_queues; i++) {
560 struct auxtrace_queue *queue = &bts->queues.queue_array[i];
561 struct intel_bts_queue *btsq = queue->priv;
562
563 if (btsq && btsq->tid == tid)
564 return intel_bts_flush_queue(btsq);
565 }
566 return 0;
567 }
568
569 static int intel_bts_process_queues(struct intel_bts *bts, u64 timestamp)
570 {
571 while (1) {
572 unsigned int queue_nr;
573 struct auxtrace_queue *queue;
574 struct intel_bts_queue *btsq;
575 u64 ts = 0;
576 int ret;
577
578 if (!bts->heap.heap_cnt)
579 return 0;
580
581 if (bts->heap.heap_array[0].ordinal > timestamp)
582 return 0;
583
584 queue_nr = bts->heap.heap_array[0].queue_nr;
585 queue = &bts->queues.queue_array[queue_nr];
586 btsq = queue->priv;
587
588 auxtrace_heap__pop(&bts->heap);
589
590 ret = intel_bts_process_queue(btsq, &ts);
591 if (ret < 0) {
592 auxtrace_heap__add(&bts->heap, queue_nr, ts);
593 return ret;
594 }
595
596 if (!ret) {
597 ret = auxtrace_heap__add(&bts->heap, queue_nr, ts);
598 if (ret < 0)
599 return ret;
600 } else {
601 btsq->on_heap = false;
602 }
603 }
604
605 return 0;
606 }
607
608 static int intel_bts_process_event(struct perf_session *session,
609 union perf_event *event,
610 struct perf_sample *sample,
611 struct perf_tool *tool)
612 {
613 struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
614 auxtrace);
615 u64 timestamp;
616 int err;
617
618 if (dump_trace)
619 return 0;
620
621 if (!tool->ordered_events) {
622 pr_err("Intel BTS requires ordered events\n");
623 return -EINVAL;
624 }
625
626 if (sample->time && sample->time != (u64)-1)
627 timestamp = perf_time_to_tsc(sample->time, &bts->tc);
628 else
629 timestamp = 0;
630
631 err = intel_bts_update_queues(bts);
632 if (err)
633 return err;
634
635 err = intel_bts_process_queues(bts, timestamp);
636 if (err)
637 return err;
638 if (event->header.type == PERF_RECORD_EXIT) {
639 err = intel_bts_process_tid_exit(bts, event->fork.tid);
640 if (err)
641 return err;
642 }
643
644 if (event->header.type == PERF_RECORD_AUX &&
645 (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) &&
646 bts->synth_opts.errors)
647 err = intel_bts_lost(bts, sample);
648
649 return err;
650 }
651
652 static int intel_bts_process_auxtrace_event(struct perf_session *session,
653 union perf_event *event,
654 struct perf_tool *tool __maybe_unused)
655 {
656 struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
657 auxtrace);
658
659 if (bts->sampling_mode)
660 return 0;
661
662 if (!bts->data_queued) {
663 struct auxtrace_buffer *buffer;
664 off_t data_offset;
665 int fd = perf_data_file__fd(session->file);
666 int err;
667
668 if (perf_data_file__is_pipe(session->file)) {
669 data_offset = 0;
670 } else {
671 data_offset = lseek(fd, 0, SEEK_CUR);
672 if (data_offset == -1)
673 return -errno;
674 }
675
676 err = auxtrace_queues__add_event(&bts->queues, session, event,
677 data_offset, &buffer);
678 if (err)
679 return err;
680
681 /* Dump here now we have copied a piped trace out of the pipe */
682 if (dump_trace) {
683 if (auxtrace_buffer__get_data(buffer, fd)) {
684 intel_bts_dump_event(bts, buffer->data,
685 buffer->size);
686 auxtrace_buffer__put_data(buffer);
687 }
688 }
689 }
690
691 return 0;
692 }
693
694 static int intel_bts_flush(struct perf_session *session,
695 struct perf_tool *tool __maybe_unused)
696 {
697 struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
698 auxtrace);
699 int ret;
700
701 if (dump_trace || bts->sampling_mode)
702 return 0;
703
704 if (!tool->ordered_events)
705 return -EINVAL;
706
707 ret = intel_bts_update_queues(bts);
708 if (ret < 0)
709 return ret;
710
711 return intel_bts_process_queues(bts, MAX_TIMESTAMP);
712 }
713
714 static void intel_bts_free_queue(void *priv)
715 {
716 struct intel_bts_queue *btsq = priv;
717
718 if (!btsq)
719 return;
720 free(btsq);
721 }
722
723 static void intel_bts_free_events(struct perf_session *session)
724 {
725 struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
726 auxtrace);
727 struct auxtrace_queues *queues = &bts->queues;
728 unsigned int i;
729
730 for (i = 0; i < queues->nr_queues; i++) {
731 intel_bts_free_queue(queues->queue_array[i].priv);
732 queues->queue_array[i].priv = NULL;
733 }
734 auxtrace_queues__free(queues);
735 }
736
737 static void intel_bts_free(struct perf_session *session)
738 {
739 struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
740 auxtrace);
741
742 auxtrace_heap__free(&bts->heap);
743 intel_bts_free_events(session);
744 session->auxtrace = NULL;
745 free(bts);
746 }
747
748 struct intel_bts_synth {
749 struct perf_tool dummy_tool;
750 struct perf_session *session;
751 };
752
753 static int intel_bts_event_synth(struct perf_tool *tool,
754 union perf_event *event,
755 struct perf_sample *sample __maybe_unused,
756 struct machine *machine __maybe_unused)
757 {
758 struct intel_bts_synth *intel_bts_synth =
759 container_of(tool, struct intel_bts_synth, dummy_tool);
760
761 return perf_session__deliver_synth_event(intel_bts_synth->session,
762 event, NULL);
763 }
764
765 static int intel_bts_synth_event(struct perf_session *session,
766 struct perf_event_attr *attr, u64 id)
767 {
768 struct intel_bts_synth intel_bts_synth;
769
770 memset(&intel_bts_synth, 0, sizeof(struct intel_bts_synth));
771 intel_bts_synth.session = session;
772
773 return perf_event__synthesize_attr(&intel_bts_synth.dummy_tool, attr, 1,
774 &id, intel_bts_event_synth);
775 }
776
777 static int intel_bts_synth_events(struct intel_bts *bts,
778 struct perf_session *session)
779 {
780 struct perf_evlist *evlist = session->evlist;
781 struct perf_evsel *evsel;
782 struct perf_event_attr attr;
783 bool found = false;
784 u64 id;
785 int err;
786
787 evlist__for_each_entry(evlist, evsel) {
788 if (evsel->attr.type == bts->pmu_type && evsel->ids) {
789 found = true;
790 break;
791 }
792 }
793
794 if (!found) {
795 pr_debug("There are no selected events with Intel BTS data\n");
796 return 0;
797 }
798
799 memset(&attr, 0, sizeof(struct perf_event_attr));
800 attr.size = sizeof(struct perf_event_attr);
801 attr.type = PERF_TYPE_HARDWARE;
802 attr.sample_type = evsel->attr.sample_type & PERF_SAMPLE_MASK;
803 attr.sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID |
804 PERF_SAMPLE_PERIOD;
805 attr.sample_type &= ~(u64)PERF_SAMPLE_TIME;
806 attr.sample_type &= ~(u64)PERF_SAMPLE_CPU;
807 attr.exclude_user = evsel->attr.exclude_user;
808 attr.exclude_kernel = evsel->attr.exclude_kernel;
809 attr.exclude_hv = evsel->attr.exclude_hv;
810 attr.exclude_host = evsel->attr.exclude_host;
811 attr.exclude_guest = evsel->attr.exclude_guest;
812 attr.sample_id_all = evsel->attr.sample_id_all;
813 attr.read_format = evsel->attr.read_format;
814
815 id = evsel->id[0] + 1000000000;
816 if (!id)
817 id = 1;
818
819 if (bts->synth_opts.branches) {
820 attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS;
821 attr.sample_period = 1;
822 attr.sample_type |= PERF_SAMPLE_ADDR;
823 pr_debug("Synthesizing 'branches' event with id %" PRIu64 " sample type %#" PRIx64 "\n",
824 id, (u64)attr.sample_type);
825 err = intel_bts_synth_event(session, &attr, id);
826 if (err) {
827 pr_err("%s: failed to synthesize 'branches' event type\n",
828 __func__);
829 return err;
830 }
831 bts->sample_branches = true;
832 bts->branches_sample_type = attr.sample_type;
833 bts->branches_id = id;
834 /*
835 * We only use sample types from PERF_SAMPLE_MASK so we can use
836 * __perf_evsel__sample_size() here.
837 */
838 bts->branches_event_size = sizeof(struct sample_event) +
839 __perf_evsel__sample_size(attr.sample_type);
840 }
841
842 bts->synth_needs_swap = evsel->needs_swap;
843
844 return 0;
845 }
846
847 static const char * const intel_bts_info_fmts[] = {
848 [INTEL_BTS_PMU_TYPE] = " PMU Type %"PRId64"\n",
849 [INTEL_BTS_TIME_SHIFT] = " Time Shift %"PRIu64"\n",
850 [INTEL_BTS_TIME_MULT] = " Time Muliplier %"PRIu64"\n",
851 [INTEL_BTS_TIME_ZERO] = " Time Zero %"PRIu64"\n",
852 [INTEL_BTS_CAP_USER_TIME_ZERO] = " Cap Time Zero %"PRId64"\n",
853 [INTEL_BTS_SNAPSHOT_MODE] = " Snapshot mode %"PRId64"\n",
854 };
855
856 static void intel_bts_print_info(u64 *arr, int start, int finish)
857 {
858 int i;
859
860 if (!dump_trace)
861 return;
862
863 for (i = start; i <= finish; i++)
864 fprintf(stdout, intel_bts_info_fmts[i], arr[i]);
865 }
866
867 u64 intel_bts_auxtrace_info_priv[INTEL_BTS_AUXTRACE_PRIV_SIZE];
868
869 int intel_bts_process_auxtrace_info(union perf_event *event,
870 struct perf_session *session)
871 {
872 struct auxtrace_info_event *auxtrace_info = &event->auxtrace_info;
873 size_t min_sz = sizeof(u64) * INTEL_BTS_SNAPSHOT_MODE;
874 struct intel_bts *bts;
875 int err;
876
877 if (auxtrace_info->header.size < sizeof(struct auxtrace_info_event) +
878 min_sz)
879 return -EINVAL;
880
881 bts = zalloc(sizeof(struct intel_bts));
882 if (!bts)
883 return -ENOMEM;
884
885 err = auxtrace_queues__init(&bts->queues);
886 if (err)
887 goto err_free;
888
889 bts->session = session;
890 bts->machine = &session->machines.host; /* No kvm support */
891 bts->auxtrace_type = auxtrace_info->type;
892 bts->pmu_type = auxtrace_info->priv[INTEL_BTS_PMU_TYPE];
893 bts->tc.time_shift = auxtrace_info->priv[INTEL_BTS_TIME_SHIFT];
894 bts->tc.time_mult = auxtrace_info->priv[INTEL_BTS_TIME_MULT];
895 bts->tc.time_zero = auxtrace_info->priv[INTEL_BTS_TIME_ZERO];
896 bts->cap_user_time_zero =
897 auxtrace_info->priv[INTEL_BTS_CAP_USER_TIME_ZERO];
898 bts->snapshot_mode = auxtrace_info->priv[INTEL_BTS_SNAPSHOT_MODE];
899
900 bts->sampling_mode = false;
901
902 bts->auxtrace.process_event = intel_bts_process_event;
903 bts->auxtrace.process_auxtrace_event = intel_bts_process_auxtrace_event;
904 bts->auxtrace.flush_events = intel_bts_flush;
905 bts->auxtrace.free_events = intel_bts_free_events;
906 bts->auxtrace.free = intel_bts_free;
907 session->auxtrace = &bts->auxtrace;
908
909 intel_bts_print_info(&auxtrace_info->priv[0], INTEL_BTS_PMU_TYPE,
910 INTEL_BTS_SNAPSHOT_MODE);
911
912 if (dump_trace)
913 return 0;
914
915 if (session->itrace_synth_opts && session->itrace_synth_opts->set) {
916 bts->synth_opts = *session->itrace_synth_opts;
917 } else {
918 itrace_synth_opts__set_default(&bts->synth_opts);
919 if (session->itrace_synth_opts)
920 bts->synth_opts.thread_stack =
921 session->itrace_synth_opts->thread_stack;
922 }
923
924 if (bts->synth_opts.calls)
925 bts->branches_filter |= PERF_IP_FLAG_CALL | PERF_IP_FLAG_ASYNC |
926 PERF_IP_FLAG_TRACE_END;
927 if (bts->synth_opts.returns)
928 bts->branches_filter |= PERF_IP_FLAG_RETURN |
929 PERF_IP_FLAG_TRACE_BEGIN;
930
931 err = intel_bts_synth_events(bts, session);
932 if (err)
933 goto err_free_queues;
934
935 err = auxtrace_queues__process_index(&bts->queues, session);
936 if (err)
937 goto err_free_queues;
938
939 if (bts->queues.populated)
940 bts->data_queued = true;
941
942 return 0;
943
944 err_free_queues:
945 auxtrace_queues__free(&bts->queues);
946 session->auxtrace = NULL;
947 err_free:
948 free(bts);
949 return err;
950 }
Linux with added FPC-III support